Talk by Chris Hillar, former Post-Doc in the Redwood Center for Theoretical Neuroscience at UC Berkeley. Given at the Redwood Center lab meeting.
Title: Elements of Theoretical Neural Science
Abstract: I
will present 4 topics from the theory of brain computation:
Memory/Encoding, Invariance, Behavioral Rhythms, and Language. Each
topic incorporates a particular mathematical model of the relevant
processing: Discrete Recurrent Neural Nets (DRNNs) / Rate-Distortion
Theory, The Bispectrum from Applied Group Theory, Renewal Processes in
Time-Series Modeling, and Hyperdimensional Computing. To appeal to the
largest audience possible, I will not give too many details of the
techniques involved, but instead focus on high-level concepts and
explicit examples. Details can be found in the references below.
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
References:
Efficient and optimal binary Hopfield associative memory storage using minimum probability flow
Robust exponential memory in Hopfield networks
A Hopfield recurrent neural network trained on natural images performs state-of-the-art image compression
Exploring discrete approaches to lossy compression schemes for natural image patches
Maximum entropy distributions on graphs
When is sparse dictionary learning well-posed?
A novel set of rotationally and translationally invariant features for images based on the non-commutative bispectrum
Informational and Causal Architecture of Discrete-Time Renewal Processes
What We Mean When We Say “What’s the Dollar of Mexico?”: Prototypes and Mapping in Concept Space
